Early-life threat and deprivation: Are children similarly affected by exposure to each?

Abstract

Extensive evidence documents negative consequences of adversity for children’s development. Here we extend such work by looking beyond average effects to consider variation in susceptibility to both threat and deprivation in terms of cognitive and social-emotional development, using an influence-statistic methodology. Data comes from the ongoing Avon Longitudinal Study of Parents and Children (N = 14,541, 49.1% female, age range: 6mo. to 12yrs., race & ethnicity: 97.8% white, .4% black, and .6% other). With respect to anticipated associations of threat with problem behavior and of deprivation with cognition, results of this pre-registered research revealed that a roughly equal proportion of children proved to be susceptible in a domain-general manner (similarly influenced) and a domain-specific one (dissimilarly influenced). Implications for intervention are considered.

Experiencing early life adversity is a well-studied and significant source of influence on children’s development (Nelson et al., 2020). Two core underlying dimensions of adversity—threat and deprivation—cover a wide range of negative experiences common in childhood. Experiences involving threat (e.g., physical, or sexual abuse), and deprivation (e.g., neglect) each have distinct influences on children’s emotional and cognitive development (McLaughlin & Sheridan, 2016; Sheridan & McLaughlin, 2014).

Apart from research informed by diathesis-stress or dual-risk thinking, which posits that some children are more vulnerable to adversity than others, virtually all research on adversity documents what are essentially average effects of adverse childhood experiences on the sample under study (Ellis, Sheridan, Belsky & McLaughlin, 2022). As such, it disregards the fact that there is variation in terms of which children are affected and the extent to which this is the case. Differential susceptibility theory posits that some children are more developmentally plastic and therefore, prove more susceptible to both negative and positive environmental effects than do others (Belsky, Bakermans-Kranenburg & van IJzendoorn, 2007; Belsky & Pluess, 2009, 2013; Ellis et al., 2011).

Yet, what has been repeatedly noted in the work just cited but only recently subject to empirical inquiry is whether developmental plasticity should be regarded as a general trait (Belsky, Zhang & Sayler, 2021). That is, are the same children affected, for better or for worse, more than others by different adverse experiences and exposures, such as threat and deprivation? This is the focus of the present report. To address this issue, the research reported herein draws on data from the Avon Longitudinal Study of Parents and Children (ALSPAC); assesses the extent to which particular children’s cognitive and social-emotional development are most and least affected by threat and deprivation; and evaluates whether children are similarly affected to each adversity. The work presented was pre-registered (https://osf.io/wpcxh) using Open Science Framework (OSF).

Threat and Deprivation Effects

Much prior research documenting the effects of early life adversity has relied on cumulative-risk approaches that sum adverse experiences regardless of type, chronicity, or severity in an effort to account for variation in children’s development (Anda et al., 2005). This body of work chronicles strong links between experiencing adversity and many negative mental health outcomes (McLaughlin et al., 2010, 2012). For example, a cumulative-risk index compromised of seven indicators—physical abuse, sexual abuse, single-parent household, number of caregiver transitions, number of school transitions, exposure to community violence, and intellectual functioning—predicted mental health symptoms among a sample of maltreated youths (aged 9-11), reliably differentiating between those children who did and did not score in the clinical range for anxiety and externalizing/internalizing problems (Raviv et al., 2010). Thus, the more adversity children experienced, the more likely they were to develop mood, anxiety, and disruptive behavior disorders (for review see Evans, Li & Whipple, 2013; Juwariah et al., 2022).

Despite such power to predict variation in development, cumulative-risk models afford little insight into the mechanisms accounting for detected links between adverse exposures and developmental outcomes (McLaughlin al., 2021). The Dimensional Model of Adversity and Psychopathology (DMAP) addressed this lacuna by distinguishing two core underlying dimensions of adversity: threat and deprivation (McLaughlin et al., 2014, 2021; McLaughlin & Sheridan, 2016). These distinct dimensions of adversity cover a wide range of negative experiences and exposures that are all too common in the childhoods of many children. Threat encompasses experiences involving actual harm or threat, and deprivation involves reductions in cognitive or social inputs. It is well appreciated that these distinct forms of adversity often co-occur, even as these dimensions can be measured separately and have unique effects on development (Ellis et al., 2020; McLaughlin et al., 2012). Because neural plasticity is heightened in early childhood and adolescence (Fandakova, 2020), adverse experiences and exposures during these developmental periods are especially likely to produce lasting changes in the brain—and thus psychological and behavioral development.

According to the DMAP, brain plasticity is considered the primary mechanism through which environmental exposures and experiences shape learning and development. This is because neural plasticity mechanisms are sensitive to specific types of environmental inputs (Takesian & Hensch, 2013; Kolb & Gibb, 2014; Ho & King, 2021), making it unlikely that the neurodevelopmental processes influenced by early life adversity that in turn influence psychological and behavioral development are similar across all forms of adverse environments. Therefore, the central tenant of the DMAP is that different dimensions of adversity will have different effects on neurodevelopment, with resulting phenotypic sequelae being at least partially distinct. Notably, evidence is consistent with this theoretical claim (Kolb & Gibb, 2014; McLaughlin & Gabard-Durnam, 2021; Nelson & Gabard-Durnam, 2020).

Disruptions in threat processing are thought to be a key neurodevelopmental mechanism underlying associations between exposure to childhood trauma and the onset of internalizing and externalizing symptoms (McLaughlin & Lambert, 2017). Specifically, DMAP predicts that experiences of threat during childhood alter developing neural networks in ways that facilitate the swift identification of danger in the environment and initiate defensive responses for safety (Sheridan & McLaughlin, 2014; McLaughlin & Lambert, 2017). It further stipulates that experiences of threat are related to problematic behavior via fear learning, emotional reactivity, and difficulties with emotion regulation (McLaughlin & Lambert, 2017). There is strong evidence to support this claim; a relevant review revealed that experiences of threat early in life are associated with changes to both the structure and function of brain regions involved in emotion learning, including reduced amygdala and hippocampal volume, as well as elevated amygdala responsiveness to threat cues (McLaughlin, Weissman & Bitran, 2019).

In contrast, children exposed to deprivation in early life do not generally exhibit similar alterations in emotional processing associated with threat. Instead, experiences of deprivation are linked to cognitive difficulties, including language ability and executive functioning. Like threat, though, deprivation is also associated with increased risk for psychopathology and difficulties in school (Miller et al., 2021; Lonigan et al., 2017). Reductions in expected environmental inputs influence future cognition by altering the foundation on which more complex forms of thinking are based. This is because the brain selectively eliminates synaptic connections that are utilized infrequently (Faust, Gunner & Schafer, 2021), a likely situation in the face of deprivation.

In fact, DMAP further predicts that exposure to environments characterized by limited social and cognitive stimulation can contribute to accelerated and extreme synaptic “pruning,” thereby, leading to reductions in the thickness and volume of cortical regions essential for the processing of social and cognitive experiences (McLaughlin et al., 2017; Sheridan & McLaughlin, 2016). Evidence supporting these claims comes from research on children who experience extreme deprivation, such as growing up in understaffed Romanian orphanages (Mackes et al., 2020), as well as those exposed to less severe forms of deprivation, including poverty and neglect (Noble et al., 2007). Children experiencing such deprived conditions exhibit reductions in grey matter volume and, presumably in consequence, difficulties in executive functioning and linguistic development.

Differential Susceptibility to Environmental Influences

Regarding the effects of two different types of adversity just reviewed, as well as of cumulative risk, there is much theory and evidence that children vary in the extent that these contextual conditions shape development. According to differential susceptibility thinking, it is not only that some children are especially vulnerable to adversity, but that children vary in their susceptibility to both positive and negative developmental experiences and environmental exposures (Belsky, Bakermans-Krannenburg & van Ijzendoorn, 2007; Belsky & Pluess, 2009, 2013; Ellis et al., 2011). Differential susceptibility to environmental influences is often conceptualized in trait-like terms, implying that children, whether proving high or low in susceptibility, develop this way across diverse contextual inputs (e.g., harsh parenting, cognitive stimulation) and developmental sequelae (e.g., aggression, executive function). Examples of trait-like terminology include “orchids and dandelions” (Boyce & Ellis, 2005) and “highly sensitive persons” (Aron & Aron, 1997), in addition to frequently reproduced graphical depictions of differential susceptibility (Belsky et al., 2007).

This domain-general view of differential susceptibility, which has even been questioned by promulgators of the differential susceptibility hypothesis (e.g., Belsky & Pluess, 2009, 2013) would seem misaligned with current neurobiological thinking which presumes that different adverse experiences influence different brain processes and thereby different developmental phenotypes (e.g., McLaughlin, Sheridan, & Lambert, 2014). Nevertheless, there are empirical grounds for taking the domain-general approach seriously. Numerous studies of differential susceptibility—whether focused on temperament, physiology, or genes as moderators of environmental effects in a for-better-or-for-worse manner—find that the same putative plasticity factors condition—in remarkably similar ways—the effects of a broad range of environmental features (e.g., prenatal stress, marital conflict, economic hardship, maternal sensitivity) on a wide variety of developmental phenotypes, including sleep, pubertal development, attentional bias, executive function and externalizing behavior (Belsky & Pluess, 2009; 2013; Ellis et al., 2011). In short, despite the seeming simplicity of a general trait-like view of susceptibility to environmental influences, there is empirical evidence to support this notion.

What is critical to understand before embracing the between-study evidence just highlighted, is that the conclusions just drawn were based on results of different studies of different children exposed to different contextual conditions. This raises the question of what happens when a within-study analysis is carried out in which the same children are considered along with their susceptibility to different environmental conditions. Does evidence of domain generality or domain specificity emerge? When a novel influence-statistic method is used to address this issue, evidence for domain specificity emerges, as well as domain generality.

Consider first research that examined effects of multiple environmental experiences and exposures (e.g., sensitive parenting, child care quality) on multiple developmental phenotypes (e.g., behavior problems, preacademic skill) in the case of the same children enrolled in the NICHD Study of Early Child Care and Youth Development (NICHD Research Network, 2003). It revealed that variation in susceptibility across the multiple effects examined proved to be normally, rather than bimodally distributed (e.g., orchids-dandelions), clearly consistent with domain-generality rather than specificity (Zhang et al., 2023).

Further evidence consistent with this conclusion emerged when the same team of investigators focused, using the same data set, on two well-documented child care effects. Results revealed that children whose cognitive-linguistic development proved most or least affected by the quality of childcare they experienced were generally not the same as those whose behavior problems proved most or least affected by the quantity (or dosage) of care they experienced (Belsky et al., 2022). When attention was turned to differential susceptibility to the effects of environment harshness and unpredictability (Zhang et al., 2021) and to parents and peers (Sayler et al., 2022), somewhat similar results emerged. Even though many children proved susceptible in a domain-specific manner, others proved susceptible in a domain-general manner.

Current Study

The research reported herein extends prior work on differential susceptibility by looking beyond average effects to consider individual differences in susceptibility to the consequences of threat and deprivation. Therefore, the primary goal is to assess the extent to which children prove similarly susceptible to effects of threat on problem behavior and deprivation on cognitive functioning. To achieve this goal, a novel influence-statistic approach is utilized that assesses individual differences in susceptibility to environmental influence that was used in the prior cited work. In this case, however, this study is the first utilizing this approach to examine differential susceptibility to threat and deprivation dimensions using data from a large-scale UK project.

The OSF pre-registration (linked here) delineated the following hypotheses: (1) that greater deprivation exposure will be associated with poorer cognitive abilities and intelligence; (2) that greater threat exposure will be associated with poorer socioemotional development (i.e., more externalizing and internalizing problems); and (3) that children who prove most and least susceptible to threat will be somewhat similarly affected by deprivation, but there will be children who prove highly susceptible to one exposure but not to the other (with regard to their respective anticipated outcomes). The current study utilizes education as a proxy measure for deprivation like many other prior large population-based longitudinal studies (e.g., Platt et al., 2018; Colich et al., 2020; Sheridan et al., 2017) because more detailed measures of deprivation such as neglect were not available. Additionally, measures of brain processes or structures were similarly not assessed in the original UK study. Because a reviewer called attention to evidence that deprivation is also related to problem behavior (e.g., Mazza et al., 2017), even if via a different brain mechanism, a non-registered analysis was added to the pre-registered plan (see Sensitivity Analysis). There we report the effect of deprivation on behavior problems (while controlling for threat) and the degree to which susceptibilities to the effect of threat and deprivation on this outcome are related to each other. Our original decision to focus on threat when predicting behavior problems is supported by results of recent meta-analysis (see Lee et al., 2024) indicating that the magnitude of associations between threat and problem behavior are larger than those with deprivation. Overall, this study should be considered a largely confirmatory effort given that it is a pre-registered report of a large, well-defined sample that involved specific predictions.

Method

Participants

Data for this report comes from the ongoing Avon Longitudinal Study of Parents and Children (ALSPAC, N = 14,541). ALSPAC was launched in the early 1990s to investigate modifiable influences on individuals’ health and development (among many other topics). Pregnant women were enrolled with estimated delivery dates between 1 April 1991 and 31 December 1992 within the Avon area of England (Boyd et al., 2013; Fraser et al., 2013). 20,248 pregnancies have been identified as being eligible and the initial number of pregnancies enrolled was 14,541. Child participants in this study were 49.1% female with an age range of 6 months to 12 years. Race and ethnicity of participants were 97.8% white, .4% black, and .6% other. In this sample, parental educational level was limited, with only 10.4% of mothers and 14.1% of fathers reporting degree-level qualifications. Mother-reported family-level income at 33 months showed that 14.4% of participants were from a low-income household (based on the Townsend index an area-based measure; Townsend, Phillimore & Beattie, 1988). At 8 months postpartum, 79.4% of mothers in this study reported their marital status as “married”. Further details about ALSPAC are found at http://www.bristol.ac.uk/alspac/. The study website contains details of all the data within a fully searchable data dictionary and variable search tool (http://www.bristol.ac.uk/alspac/researchers/our-data/). Ethical approval for this study was obtained from the ALSPAC Ethics and Law Committee and the Local Research Ethics Committees. Informed consent for the use of data collected via questionnaires and clinics was obtained from participants following the recommendations of the ALSPAC Ethics and Law Committee at the time.

Sample Size

The analytic sample size is based on the total ALSPAC sample of 14,541 children, minus cases that met exclusionary criteria (see Data Exclusion), as multiple imputation is used to estimate all missing data for variables included in the current report.

Data Exclusion

Children with scores of mother-reported poor health within the past year from 18 months to 3.5 years and those whose teachers reported that the child had physical disabilities, medical conditions, or developmental delay in grades 3 and 6 were excluded from the analysis. Excluded cases based on the mother and teacher-reported health indices resulted in a final analytic sample size of 14,029 children. These children were excluded because poor health, physical disabilities, medical conditions, and developmental delays may confound the influence of deprivation on cognitive deficits which is an outcome of interest in this study.

Measures

Predictor and outcome variables, as well as covariates, are delineated below. All variables were scaled so that a higher score reflects greater threat, deprivation, problem behavior, or cognitive deficits (i.e., reverse scaling as needed).

Environmental Predictors

Two sets of variables—one for threat and the other for deprivation exposures—served as predictors.

Deprivation.

Four indices of deprivation were used: (i) For frequency of visiting places we relied on mothers’ reports of cognitively stimulating outings with their children at 6 months, 18 months, 2.5 years and 3 years, such as taking their child to local shops, supermarkets, parks, and homes of friends/family (5 items total; range = 1 – 3.75); answers were given on a 5-point scale from “more than once a week” to “never”. (ii) (Lack of) parental education was self-reported by mothers and fathers separately at 32 weeks’ gestation reflecting their highest level of educational attainment (2 items total; range = 0 – 4). Answers included “none”, “certificate of secondary education”, “O level”, “A level”, “vocational” and “degree”. (iii) Aspects of parental deprivation were reported by mothers at 2 years and 3 years such as how often they allow child to play with paints/messy objects, use objects to build towers, take them to the park or playground, allow child to make a lot of noise while playing, share a meal with child, and praise or kiss child. (9 items total: range = −8 – 1). Answers were given on a 5-point scale and ranged from “every day” to “never’. (iiii) Lastly, to assess (lack of) maternal cognitive stimulation we relied on maternal reports at 6 months, 18 months, 3 years, and 3.5 years. At 6 months, 18 months and 3 years, mothers reported how often they sing to child, show/read picture books, and play with child (3 items total; range = 1 – 3.89). Answers were given on a 5-point scale ranging from “every day” to “never”. At 3.5 years, mothers were asked slightly different questions focused on different activities, including singing and reading to child, and playing with toys and other fun activities. Answers were given on a 5-point scale from “Nearly every day” to “Never. All items for each of the four indices were averaged based on mean scores. It should be noted that there is some overlap between frequency of visiting places and parental deprivation in that both indices surveyed parents regarding the frequency that they take their child to the park or playground. This overlap is discussed further in the limitations section of the Discussion.

Threat.

Four indices of threat exposure were used: (i) Neighborhood threat was based on mothers’ reports on the frequency of burglary, attacks, vandalism, and problems with youth in neighborhood at 1.75 and 2.75 years. The score was computed as the mean on 11 items (range = 0 – 2), each rated on a 3-point scale (0 = no problem or opinion, 1 = minor problem, 2 = serious problem). (ii) To assess exposure to parental domestic violence, mothers were asked when their child was 1.75 years old if they or their partner had “hit or slapped their partner in the past 3 months” and if they “threw something in anger within the past 3 months”. Answers included “yes mum did”, “yes partner did”, “yes both did” or “no not at all” and “no partner” (2 items total; range = 0 – 3). (iii) Mothers were also asked about child physical abuse by another person at 6, 12, and 18 months, and 3 and 5 years (5 items total; range = 1 – 2). Answers included “Yes” or “No”. (iiii) Lastly, aspects of parental threat were reported by mothers at 2 years. Mothers were asked, “When you’re at home with your child how often do you do the following: shout at them and slap them” (2 items total; range = 1 – 5). Answers were given on a 5-point scale ranging from “Every day” to “Never”. All items for each of the four indices were averaged based on mean scores.

Developmental Outcomes

Two sets of variables relating to children’s cognition and behavior served as outcomes.

Cognitive Deficits.

Two indices of cognition were used: (i) To assess intelligence, children completed a clinic-based assessment at age 8 using a short form of the Wechsler Intelligence Scale for Children- III (WISC; Woogler, 2001). This included both verbal and performance IQ. (ii) To assess learning, speech, and language difficulties teachers provided information in grades 3 and 6 by endorsing or not the existence of such difficulties. Answers included “Yes”, “In the past, not now” and “No”.

Problem Behavior.

Two indices of children’s behavior were used. To assess children’s emotions and behavior, teachers in grades 3 and 6 completed the short form of the Strengths and Difficulties Questionnaire (Goodman, 1997) which yielded sub-scores for (i) internalizing problems (emotional symptoms, peer problems) and (ii) externalizing problems (conduct problems, hyperactivity/inattention).

Covariates

Mothers reported on their child’s sex and race/ethnicity (e.g., white, or other race) upon enrollment, as well as their child’s exact age at the scheduled 24-month assessment. Race/ethnicity and child sex were identified as potential confounders for both childhood conditions and outcomes that could affect susceptibility scores. Income was not included as a covariate because a measure of family socioeconomic status (SES) was already included in all the models (parental education). Therefore, there would be no reason to include a second SES measure such as income that is highly collinear with parental education. Child sex and race/ethnicity covariates were coded for analysis as female = 1, male = 0, and white = 0, black = 1, and other race = 1.

Analysis Plan

Multiple steps in the analysis are organized in terms of preliminary and primary analyses. A sensitivity analysis is also described, though it was not cast as such in the pre-registration, even though the statistical analysis was specified. Here we delineate the original plan, which was carried out as stipulated.

Preliminary Analyses

The first step of the preliminary analyses involved multiple imputation of missing data using Multivariate Imputation by Chained Equations (MICE; van Buuren & Groothuis-Oudshoorn, 2011) in R; this involved all listed study variables. All subsequent preliminary, primary, and sensitivity analyses were then conducted based on 20 imputed data sets. Two data-reduction-oriented principal components analyses, one including threat and the other deprivation indicators, were conducted to limit statistical testing in the primary analysis. Pre-registration neglected to stipulate that this step would be conducted on outcomes as well as predictors. Results of principal components analyses provided the basis for creating single composite scores for each predictor and outcome. The final preliminary analyses relied on two least squares regression analyses, one using threat to predict problem behavior and the other using deprivation to predict cognitive deficits, after controlling for covariates. The decision to focus on only these associations was based on prior research reviewed earlier showing these distinctive and discriminating links between specific dimensions of adversity and their particular sequelae.

Primary Analyses

Following these preliminary analyses, primary analyses were carried out using influence statistics to assess the degree to which each child appeared susceptible to consequences of threat and deprivation documented in the prior regression analyses so that the two resulting susceptibility scores could be associated with each other; this would determine the extent to which children affected more or less by one experience were or were not affected by the other in a similar way. We relied on an influence statistic, DFBETAS, a continuous and standardized index assigned to each and every observation.

DFBETAS reflects the degree and direction of change of the regression coefficient after removing a single observation. Therefore, DFBETAS is calculated using a “leave-one-out” approach by re-estimating an association repeatedly, each time dropping a single case to measure how much such (minor) sample modification causes the full-sample association to increase (i.e., a negative influencer) or decrease (i.e., a positive influencer), usually ever so modestly. The resultant change of the slope parameters attributed to each observation for the association of interest (i.e., threat: problematic behavior; deprivation: cognitive deficits) indicates how—and the extent to which—particular individuals affect the full-sample estimate of the association (e.g., Belsley, Kuh, & Welsch, 1980; Cook & Weisberg, 1982). More concretely, the more attenuated the association becomes when a case is dropped, the more susceptible the child is to the effect in question. Conversely, the more positive the association becomes when a case is dropped, the less susceptible the child is to the effect.

The final step in the primary analysis evaluated whether children most and least affected by one of the predictor effects under investigation (i.e., threat, deprivation) were similarly affected by the other. This involved correlating the DFBETAS’ threat and deprivation susceptibility scores and cross-tabulating tercile splits of the two DFBETAS’ distributions. No significance test will be applied to the latter because it is not independent of the former.

Sensitivity Analysis

Sensitivity analysis involved first repeating the preliminary regression analysis, followed by a repeat of the primary susceptibility analysis. But this time the preliminary regression analysis included both predictors, threat and deprivation, when predicting each of the two outcomes so that only the unique variance explained by each predictor would be attributed to it. Thus, for example, when evaluating the effect of threat on the problem-behavior outcome, deprivation is treated as a covariate and when treating cognitive-linguistic functioning as the outcome, treating threat as a covariate. Recall that while the modified regression analyses (and resulting susceptibility scores) were preregistered, they were not framed as sensitivity analyses. The expectation was that the two effects in question would become more independent relative to the primary analysis because including a second adversity dimension in the regression analysis would reduce multicollinearity.

Results

Preliminary Analyses

Descriptive statistics of threat and deprivation indicators are provided in Table 1. Table 2 presents results of the (separate) principal component analyses of predictor and outcome variables. In each case, we created unit-value-based summary scores for all variables loading on the first principal component. These are labeled threat, deprivation, cognitive deficits, and problem behavior.

Table 1.

Descriptive Statistics of Threat and Deprivation Indicators

Threat Indicators	Mean	Minimum	Maximum	Standard Deviation
Parental Domestic Violence	.11	0	3	.37
Neighborhood Threat	.44	0	2	.38
Child Physical Abuse	1.03	1	2	.11
Parental Threat	3.15	1	5	.75
Deprivation Indicators	Mean	Minimum	Maximum	Standard Deviation
Parental Cog. Stimulation	1.35	1	3.89	.35
Parental Deprivation	−6	−8	1	.97
Frequency of Visiting Places	1.81	1	3.75	.26
Parental Education	1.85	0	4	1.11

Notes: Cog. = Cognitive. Higher scores reflect more deprivation/threat.

Table 2.

Principal Component Weightings of Data-Reduction Analyses: (A) Threat Predictors, (B) Deprivation Predictors, (C) Behavioral Outcomes and (D) Cognitive Outcomes^*

A. Threat Predictors				B. Deprivation Predictors
Loadings				Loadings
	Factor 1				Factor 1	Factor 2
Parental Domestic Violence	.59			Parental Cognitive Stimulation	.80	.13
Neighborhood Threat	.56			Parental Deprivation	.76	−.05
Child Physical Abuse	.51			Frequency of Visiting Places	.54	−.58
Parental Threat	.49			Parental Education	.29	.85
Eigenvalue	1.16			Eigenvalue	1.59	1.07
Variance	29.04%			Variance	39.66%	26.88%
C. Behavioral Outcomes				D. Cognitive Outcomes
Loadings				Loadings
	Factor 1	Factor 2	Factor 3		Factor 1	Factor 2
Hyperactivity Grade 6	.63	−.17	−.35	Verbal IQ	.63	−.48
Conduct Prob. Grade 6	.60	−.27	−.31	Performance IQ	.57	−.55
Hyperactivity Grade 3	.58	.36	−.18	Learning Diff. Grade 6	.55	.29
Conduct Prob. Grade 3	.54	.41	−.15	Learning Diff. Grade 3	.54	.31
Emotional Symp. Grade 6	.48	−.49	.26	Speech/Language Grade 3	.36	.39
Peer Problems Grade 6	.48	−.49	.26	Speech/Language Grade 6	.30	.48
Peer Problems Grade 3	.42	.40	.40
Emotional Symp. Grade 3	.23	.31	.63
Eigenvalue	1.95	1.17	1.13	Eigenvalue	1.54	1.11
Variance	24.34%	14.65%	14.07%	Variance	25.64%	18.43%

Notes: Conduct Prob. = conduct problems; Emotional Symp. = emotional symptoms; Learning diff. = learning difficulties; Speech/Language = speech and language difficulties.

^*Only factors with eigenvalues equal to or greater than 1.0 are displayed.

Intercorrelations of these composite measures are displayed in Table 3, an analysis not specified in the pre-registration of the study. These correlations, even though modest, provide validation of the factor-derived constructs specific to this investigation in that most associations are consistent with expectations. Specifically, greater deprivation is significantly associated with greater cognitive deficits whereas threat and deprivation were similarly associated with problem behavior. Notably, deprivation predicts cognitive deficits more strongly than problem behavior, as well as more strongly than threat. Also worth mentioning is that threat and deprivation are positively associated. These findings are consistent with observations made in the Introduction regarding deprivation predicting cognitive deficits but, not those relating to threat predicting problem behaviors.

Table 3.

Inter-correlation of Constructs

Composite	Threat	Deprivation	Problem Behavior
Threat	-
Deprivation	.11**	-
Problem Behavior	.13**	.14**	-
Cognitive Deficits	.08**	.22**	.29**

Notes:

^**= p < .01.

The final preliminary analysis involved evaluating—in two separate exploratory regression models on which DFBETAS susceptibility scores would be based—whether, net of covariates, greater threat was associated with greater behavior problems and whether greater deprivation was associated with more cognitive deficits. Inspection of Table 4 reveals that these expectations were confirmed.

Table 4.

Regression Results of (A) Threat Effects on Problem Behavior and (B) Deprivation Effects on Cognitive Deficits.

A. Threat Effect Model
Predictor Variables	B	SE B	t	df	p
Threat	.10	.02	4.20	140.91	<.001
Child Sex	−.60	.02	−27.84	478.59	<.001
Child Race/Ethnicity	.05	.04	1.56	59.97	.12
Age	.04	.01	2.59	40.94	<.05
B. Deprivation Effect Model
Predictor Variables	B	SE B	t	df	p
Deprivation	.04	.01	7.16	755.09	<.001
Child Sex	−.12	.01	−9.64	1681.05	<.001
Child Race/Ethnicity	.03	.02	1.60	78.83	.11
Age	.02	.01	2.83	63.87	<.01

Primary Analyses

The first step in the primary analyses involved calculating, by means of DFBETAS, the degree of susceptibility of each child to the effects of threat on problem behavior and deprivation on cognitive deficits. Recall that this was accomplished via the leave-one-out procedure and thus re-running the two whole-sample regression analyses 14,029 times, once using the threat composite to predict problem behavior and once the deprivation composite to predict cognitive deficits. With individual susceptibility scores estimated for each case, the resulting DFBETAS for threat and deprivation were correlated. The correlation between the two influence-statistic-derived susceptibility scores was significant but small (r = .07, p < .0001), indicating that children most or least susceptible to the adverse effects of threat on problem behavior proved somewhat similarly susceptible to the adverse effects of deprivation on cognitive deficits. The fact that a small effect proved so statistically reliable no doubt is a result of the very large sample.

Pre-registered exploratory efforts were undertaken to provide further descriptive insight into the association between the two susceptibility scores, dividing each DFBETAS’ distribution, arbitrarily, into thirds and cross tabulating them. Inspection of the diagonal in Table 5 running from the bottom-right corner to the top-left corner indicates that a bit more than a third of the children (36.8%) scored in the same tercile of both susceptibility distributions (i.e., low, moderate, high), thus appeared to be influenced in a relatively domain-general manner. Having said that, only a quarter of the sample proved highly susceptible to the effects of both threat and deprivation (12.7%) or highly unsusceptible to both of these effects (12%).

Table 5.

Tercile Splits of Susceptibility to Threat and Deprivation Effects (based on DFBETAS)

	Susceptibility to Threat
Susceptibility to Deprivation	Low	Moderate	High	Total
Low	1695	1542	1439	4676
Moderate	1540	1681	1456	4677
High	1441	1454	1781	4676
Total	4676	4677	4676	14,029

Further inspection of cells in Table 5 indicates that 10.3% of the total sample proved highly susceptible to threat but highly unsusceptible to deprivation, with the comparable figure for the reverse configuration also being 10.3%. Thus, a fifth of the sample proved highly susceptible to one experience and highly unsusceptible to the other, thus appearing to be susceptible in a more domain-specific manner. In other words, only a little more of the sample proved similarly susceptible to both effects (i.e., ~25%) than did those who proved very different in their susceptibility to the two effects (i.e., ~20%).

Sensitivity Analysis

The final set of analyses re-evaluated consistency in individual differences in susceptibility to threat and deprivation exposures by first including, in a revised version of the preliminary regression analysis, both predictors in a single prediction model (along with covariates) before calculating the intercorrelation of susceptibility scores (i.e., DFBETAS). Recall that it was expected that this would yield even more evidence of inconsistency in susceptibility to the two effects under consideration—because the variance that threat and deprivation shared would be attributed to neither, thus making their effects more independent. Results proved consistent with this domain-specific expectation in that the two derived susceptibility effects (e.g., for threat and deprivation) proved somewhat less correlated (r = .01, p < .001) than in the primary analysis (r = .07, p<.001), with this difference proving statistically significant (z = 5.03, p < .001).

Finally, we conducted a non-registered analysis to examine the effect of deprivation on problem behavior, controlling for threat. Results revealed that deprivation more strongly predicted behavior problems with threat controlled (beta = .11, p < .001), than did threat with deprivation controlled (beta = .07, p < .001).

Discussion

Recall that the primary goal of the work reported was to look beyond average effects of two distinct dimensions of adversity, threat and deprivation, on, respectively, two different developmental sequelae, problem behavior and cognitive deficits. Toward this end, we again relied on an influence-statistic methodology to estimate the nature and magnitude of each of the two effects on individual children in order to evaluate whether children proved to be similarly or differently affected by the two sources of influence in question. Results of the preliminary analysis revealed that some of our initial expectations outlined in the preregistration were confirmed in that greater deprivation predicted more cognitive deficits but, surprisingly, greater threat did not predict problem behavior. Moreover, non-preregistered sensitivity analysis revealed that deprivation predicted behavior problems when controlling for threat somewhat more strongly than did threat when deprivation was controlled.

Clearly, some of the preliminary findings just summarized proved contrary to our expectations. In fact, these results did not align with the findings from the recent meta-analysis by Lee and colleagues (2024) that showed the effect sizes between threat and psychopathology are larger than those with deprivation. One possible explanation for the inconsistency in question is that variability in measurement approaches—such as different informants or different operationalizations of adversity dimensions—can affect whether and the extent to which adversity is predictive of youth psychopathology (Lee et al., 2024). For example, in the current study our problem behavior outcome was limited to teacher-report only; quite conceivably results might have been different—and more in line with those of the meta-analysis—had we focused on youth or parent reports. It needs to be recognized as well that prior research indicates that poverty—which is surely a form of deprivation—is predictive of problem behavior, (Mazza et al., 2017; Fitzsimons et al., 2017), thereby indicating that the latter is not only a product of threatening experiences. Also worth considering is evidence like our own showing that experiences of threat and deprivation are somewhat differentially associated with different developmental phenotypes in children (Usacheva et al., 2022; Schäfer et al., 2023; Vogel et al., 2021; Machlin et al., 2019), in all likelihood because they have distinct influences on neurobiology, specifically differences in brain white matter within the cingulum and the uncinate fasciculus (Banihashemi et al., 2021), which, of course, went unmeasured in the research reported herein. One last point to consider is that this report draws on data from a population study, not one, like many others included in the meta-analysis focused principally on extreme adversity such as poverty (e.g., Bachman et al., 2022) and institutionalization (e.g., Zhang et al., 2019; Henry et al., 2021) or psychopathology, with the latter measured using clinical interview by trained caseworkers (Henry et al., 2021) .In sum, while we cannot be certain what accounts for the inconsistency between our preliminary findings and meta-analytic evidence, there are a number of reasons that could account for this fact.

Now turning to our primary results, findings indicated that children most or least susceptible to the adverse effects of threat on problem behavior proved somewhat similarly susceptible to the adverse effects of deprivation on cognitive deficits. To provide further descriptive insight into this association, pre-registered exploratory efforts made clear that more than a third of the children scored in the same tercile of both susceptibility distributions (i.e., low, moderate, high). In fact, a full quarter of the sample proved highly susceptible to the effects of both threat and deprivation or highly unsusceptible to both of these sources of influence. Just as notable, however, is that further consideration of categorical tercile splits revealed that a fifth of the sample proved highly susceptible to one exposure and highly unsusceptible to the other. In sum, only a little more of the sample proved similarly susceptible—whether high or low—to both dimensional effects (i.e., ~25) than did those who proved very different in their susceptibility to the two effects (i.e., ~20%). In other words, whereas some children proved to be influenced in a domain-general way by the two effects under investigation, others appeared to be affected in a domain-specific manner. Such variation in children’s susceptibility to environmental influences has never before been considered, much less documented, in research on threat and deprivation.

The findings of the primary analysis are consistent with other recent investigations of differential susceptibility to environmental influences using the same influence-statistic approach (Zhang, Widaman, & Belsky, 2021). This prior work also chronicled both domain-generality and domain-specificity among different children with respect to child care effects (Belsky et. al., 2022), effects of environmental harshness and unpredictability (Zhang et al., 2021), adversity exposure in early childhood and adolescence (Belsky & Andersen, 2022), and parent and peer effects (Sayler, et al., 2022).

The results suggest that some children are more generally susceptible to environmental effects, others are generally unsusceptible, and most fall somewhere in between. Just as notable, however, is that a separate report considering multiple family and child care predictors of multiple developmental phenotypes yielded evidence more consistent with domain specificity than domain generality (Zhang, Widaman & Belsky, 2023). The same was true, of course, once only the unique effects of threat and deprivation on, respectively, problem behavior and cognitive-linguistic development were examined in the sensitivity analysis in the current study. Collectively, these results seem in line with evolutionary analysis stipulating that it would have been beneficial—in terms of reproductive fitness—throughout human evolutionary history for individuals to vary in their susceptibility to specific experiences and exposures (Belsky, 2005; Belsky & Pluess, 2009; Ellis et al., 2011).

The results of this inquiry do not address a most important question: What characteristics of individuals make some more and others less susceptible to particular environmental effects on particular developmental outcomes. Is it their temperaments as young children, physiological reactivity, or genetics, all of which have been highlighted as potential “plasticity factors” in the differential susceptibility literature (Belsky & Pluess, 2009, 2013; Ellis et al., 2011), or some other source of influence? This empirical question was not pursued in the current inquiry because of the absence of strong theory to guide this effort, and it seemed like the time had passed to simply conduct exploratory work checking out a variety of seemingly alternative hypotheses (Belsky et al., 2021; Zhang et al. 2022). More theory and research are called for when thinking about plasticity factors.

This study has several strengths. First, by utilizing a dimensional approach to studying adversity, it was possible to distinguish differential associations of distinct experiences, threat, and deprivation with, respectively, cognitive deficits and problem behavior. Second, reliance on the influence statistic DFBETAS once again enabled the evaluation of the degree to which individual children appeared susceptible to the adversity effects investigated in order to determine the extent to which children affected by one experience were or were not affected by the other in a similar way and to a similar extent. Lastly, this study benefitted from its large sample size and longitudinal design. Recall, as well that it was preregistered.

Some limitations should also be noted. The first issue was our reliance on parental report for the threat indicators rather than more objective measurements. Importantly, it should also be noted that our deprivation indicator is mostly a measure of the frequency of exposure to stimulating activities with their mother (e.g., singing to child, playing with child, visiting places) and does not assess the quality of the interaction or mother’s behavior—such as sensitive responsiveness, disregard or intrusiveness—apart from one item that asks the frequency of praising or kissing their child. Given the population nature of ALSPAC, it is perhaps not all that surprising that measurement was somewhat impoverished. Our approach, however, is not unlike most other studies of deprivation that examine variation in cognitive stimulation—as many include both parental behaviors and engagement in certain activities (e.g., sharing meals). An example of this would be the Home Observation for Measurement of the Environment (HOME) instrument which can be used to measure the quality of cognitive stimulation and emotional support provided by the child’s family (Bradley & Caldwell, 1977). Another limitation was the overlap between deprivation indicators frequency of visiting places and parental deprivation such that both items indices measured how often mothers and their children visited the playground or park which may have affected our measurement of deprivation.

Additionally, our study used parental education as a proxy measure for deprivation. As noted previously, many prior large population-based longitudinal studies similar to ours do not typically include detailed measures of deprivation such as cognitive stimulation or neglect but do include parental education as an indicator of deprivation (e.g., Platt et al., 2018; Colich et al., 2020; Sheridan et al., 2017). We acknowledge that parental education is a proxy measure that does not capture deprivation as well as other measures. It is, however, a widely measured construct. Future studies should include more detailed and direct measures of cognitive stimulation, both within and outside the home, to allow for better testing of threat and deprivation associations with developmental outcomes. Despite observational measures being considered the gold standard for deprivation, there are also well-validated parent-report measures that can be used in large-scale studies such as the conflict tactic scale (parent-child) and the multidimensional neglectful behavior scale (see Berman et al., 2022). In our study, parental education also did not align strongly with other indicators of deprivation, but this was not a concern as we approached this issue conceptually not just statistically given that prior work has used education as an indicator of deprivation. Statistically, latent factors are not an appropriate way of modeling dimensions of adversity. This is because dimensional models do not presume that threat necessarily co-occur more with other threats or that deprivation co-occurs necessarily with other forms of deprivation. Mechanisms and developmental processes influenced by threat exposures are similar and at least partially distinct from the same associated with deprivation, a point made clear in a recent commentary about reliance on latent factors to model these two dimensions of adversity (McLaughlin, Weissman and Flournoy, 2023).

Another limitation of the work presented herein is that most indices of deprivation and threat were largely in the mild to moderate range. Considering that the purpose of dimensional models is to capture adverse experiences across a continuum of severity and frequency rather than focusing only on the highest severity cases (as has typically been done in prior work with samples of children raised in orphanages or in the child welfare system), the distribution that we see here is expected given the that the Avon Longitudinal Study is population representative. The implication of this is that we captured associations that likely reflect those in the broader population of children (thus increasing generalizability of findings), and that these associations are smaller in magnitude than would be expected in samples focusing on children exposed to more extreme and far less common forms of deprivation and adversity.

Moreover, while the study is longitudinal in the temporal ordering of predictors and outcomes, the fact that we have combined predictors measured at different time points, as done with outcome measurements, means we have not fully taken into account the temporal ordering of the data. Additionally, the analytic approach doesn’t completely account for measurement error in the susceptibility scores. Composite scores were used in this report to index the environmental predictors, outcomes, and the susceptibility measure. Therefore, at least to some degree, it is likely that the influence-statistic estimates of susceptibility reflect measurement error rather than a precise assessment of such variation. Finally, a clear limitation is the majority white (i.e., ~ 98%) sample, as the findings from this study may not be generalizable to other races/ethnicities or even necessarily different geographic locales.

The results of this study also carry potential implications for intervention. Some children in our study proved more susceptible to certain environmental exposures than to others (i.e., threat and deprivation). This suggests that intervention efficacy would likely vary across individuals, as it is known to do (Belsky & van Ijzendoorn, 2015). After all, heterogeneous effects are quite common to virtually all interventions, whether targeting children, parents, or other adults. It seems eminently possible then that, for example, interventions that address behavioral problems may be more effective with children with a history of threat-related experiences (e.g., physical abuse) and susceptibility to them. In contrast, interventions that target executive functioning or language skills may be more appropriate for children who proved more susceptible to deprivation effects (e.g., neglect).

Conclusion

Early-life adversity—specifically threat and deprivation experiences—have well-documented negative effects on children’s cognitive and behavioral development. The findings presented herein extend prior research on dimensions of adversity and on differential susceptibility by investigating whether children are similarly affected by threat and deprivation. Results revealed that with respect to documented associations of threat with problem behavior and deprivation with cognitive deficits, a roughly equal proportion of children proved to be similarly influenced (i.e., domain-general) and dissimilarly influenced (i.e., domain-specific). The former proved less and the latter more so when only the unique effects of the two sources of influence were considered. In either case, our findings contribute to work on dimensional approaches to studying adversity as it was possible to distinguish differential associations of distinct threat and deprivation experiences on cognitive and behavioral outcomes which could be beneficial for future interventions.